AI Agent Rule Generator

Pick your coding agent, pick the things it should never be allowed to do, and get a working permission file in the format that agent actually parses. No account, no install, no sales call.

116 rules across 10 packs Vendor-native output, not a generic template Every rule tells you what it cannot cover

Choose your agent

Each of these reads a different file, in a different format. The generator emits the one your agent actually loads.

Choose your rule packs

Four packs are preselected as a sensible baseline. A pack turns on every rule inside it, and you can untick individual rules below.

Review the individual rules

This is the part worth reading. A rule that blocks something your build genuinely needs is a rule you should untick now rather than debug later.

settings.json 0 / 50 rules

Name and work email to unlock. No card, no call.

What this file cannot do

These rules match text. They hold until somebody writes the same command a different way: curl x | bash becomes curl -o /tmp/x x && sh /tmp/x, and a path rule does not see a Python script that opens the file itself.

That is not our opinion. Anthropic documents that Bash patterns constraining arguments are fragile, Cursor states its allowlist is not an absolute security boundary, and Codex treats bash -lc "..." as one opaque invocation.

A generated file is still worth having: it closes the obvious paths and it is a statement of intent your team can read. Just know what it is.